1440-1500 20171002 Tantsura The Critical Role v1 · PDF...

$Page 1: 1440-1500 20171002 Tantsura The Critical Role v1 · PDF file@ABCDEFGHIJKLMNOPQRSTUVWXYZ[\]^_`abcdefghijklmnopqrstuvwxyz{|}~ ... In fact, a path is directly usable by any router; no$
The cri(cal role of Maximum SID Depth (MSD) hardware limita(ons in Segment Rou(ng ecosystem and how to work around those

Jeff Tantsura IAB member, IETF RTGWG Chair VP Network Architecture, Futurewei, Future Networks

Guedrez Rabah PhD candidate at Orange Labs Orange Labs Lannion

NANOG71, October 2017 San Jose, CA

Slide title 44 pt

Text and bullet level 1 minimum 24 pt

Bullets level 2-5 minimum 20 pt

Characters for Embedded font: !"#$%&'()*+,-./0123456789:;<=>?@ABCDEFGHIJKLMNOPQRSTUVWXYZ[\]^_`abcdefghijklmnopqrstuvwxyz{|}~¡¢£¤¥¦§¨©ª«¬®¯°±²³´¶·¸¹º»¼½ÀÁÂÃÄÅÆÇÈËÌÍÎÏÐÑÒÓÔÕÖ×ØÙÚÛÜÝÞßàáâãäåæçèéêëìíîïðñòóôõö÷øùúûüýþÿĀāĂăąĆćĊċČĎďĐđĒĖėĘęĚěĞğĠġĢģĪīĮįİıĶķĹĺĻļĽľŁłŃńŅņŇňŌŐőŒœŔŕŖŗŘřŚśŞşŠšŢţŤťŪūŮůŰűŲųŴŵŶŷŸŹźŻżŽžƒȘșˆˇ˘˙˚˛˜˝ẀẁẃẄẅỲỳ–—‘’‚“”„†‡•…‰‹›⁄€™ĀĀĂĂĄĄĆĆĊĊČČĎĎĐĐĒĒĖĖĘĘĚĚĞĞĠĠĢĢĪĪĮĮİĶĶĹĹĻĻĽĽŃŃŅŅŇŇŌŌŐŐŔŔŖŖŘŘŚŚŞŞŢŢŤŤŪŪŮŮŰŰŲŲŴŴŶŶŹŹŻŻȘș−≤≥fifl

ΆΈΉΊΌΎΏΐΑΒΓΕΖΗΘΙΚΛΜΝΞΟΠΡΣΤΥΦΧΨΪΫΆΈΉΊΰαβγδεζηθικλνξορςΣΤΥΦΧΨΩΪΫΌΎΏ

ЁЂЃЄЅІЇЈЉЊЋЌЎЏАБВГДЕЖЗИЙКЛМНОПРСТУФХЦЧШЩЪЫЬЭЮЯАБВГДЕЖЗИЙКЛМНОПРСТУФХЦЧШЩЪЫЬЭЮЯЁЂЃЄЅІЇЈЉЊЋЌЎЏѢѢѲѲѴѴҐҐəәǽẀẁẂẃẄẅỲỳ№

Do not add objects or text in the

footer area

Segment Routing part 1 | Jeff Tantsura | Page 2

› Segment Routing is becoming mainstream technology, like every other new technology it comes with its own set of limitations.

› Knowing these limitations and understanding how to work around (permanently or temporarily, till new, more capable HW is available) will help you to apply the new technology in your network as well as to make right choices when choosing the next generation HW.

› Ongoing networking disaggregation, where SW and HW might come from different vendors would require additional API’s from HW vendors, MSD is one of them.

Why should you pay attention?

Slide title 44 pt







footer area


› MSD - Maximum SID Depth (defined in IETF OSPF/ISIS/BGP-LS/PCEP drafts) – Generic concept defining number of SID’s, HW/SW are capable of imposing on a given node – Applicable to both, SR-MPLS (labels) and SRv6 (SRH’s) data planes

›  Focus of this presentation is SR-MPLS data plane

› SR-MPLS – Segment Routing with MPLS data plane(defined in IETF SPRING drafts) – SID instantiated as an MPLS label, context’s set by the label value – Path (LSP/tunnel) is usually computed by a centralized entity, commonly known as PCE/SDNc – PCEP with SR extensions is a commonly used protocol to communicate the path to the ingress – MPLS label stack defines at the ingress the path a packet will take thru the network – Other actions could be defined and applied as packet traverses the network:

›  Apply a service ›  Treat a packet in a special way ›  Set context › …

Vocabulary

Slide title 44 pt







footer area


› Prefix SID – Uses SR Global Block (SRGB), must be unique within the SR domain – SRGB(‘s) is advertised by an IGP – Prefix-SID can be configured as an absolute value or an index (base+offset)

› Node SID – Node SID is a prefix SID with ‘N’ (node) bit set, it is associated with a host prefix (/32 or /128)

that identifies the node, more than 1 Node SID’s per node can be configured (think router-id)

Short SR-MPLS recap – SID types

Slide title 44 pt







footer area


› Adjacency SID – Locally significant in most implementations– can be made globally significant thru ‘L’ flag – Identifies unidirectional adjacency – In most implementations automatically allocated for each adjacency – Always encoded as an absolute (not indexed) value

› Binding SID – Can be originated by any SR capable device in the SR domain – Can be used to instantiate a new label stack at the SID originating node (anchor), hence

splitting end2end path into number of sub-paths

Short SR-MPLS recap – SID types

Slide title 44 pt







footer area


Controller

MD-SAL

BGP-LS-SR

RESTCONF

/operational

BGP-‐RIB

/operational

Topology

BGP-LS Topology Exporter

MD-SAL Set

MD-SAL Notification MD-SAL Set MD-SAL Get

Application (Topology Consumer)

BGP-LS-SR speaker (RR)

SR in SDN world – topology acquisition

node/link MSD

Slide title 44 pt







footer area


Controller

MD-SAL

PCEP SR

RESTCONF

/operational /operational

Topology

PCE

MD-SAL Set

MD-SAL Notification MD-SAL Set MD-SAL Get

Applications (topology/constrains/feedback)

PCEP SR speaker (ingress)

H-PCE

I2RS/OF/other SBI’s

SR in SDN world – SID stack provisioning

node/link MSD

Slide title 44 pt







footer area


› MSD supported by different HW/SW differs widely : – Linux (kernel 4.10): 2 SID’s, some improvements recently (as of 4.11) – Low end off the shelf (merchant) silicon, e.g. BCM Trident2: 3-5 SID’s – High end off the shelf (merchant) silicon, e.g BCM Jericho1 : 4-7 SID’s – Vendor’ silicon, e.g. Juniper’ Trio, Nokia’ FP3: 4-10+ SID’s

›  If SID stack > MSD at ingress node – Best case:

›  Service can’t be provided – Worse case:

›  Packet will get dropped somewhere in the network

What’s the problem?

Slide title 44 pt







footer area


› Source routing along an explicit labeled path

A path with Adjacency SID’s (strict encoding) MSD = 5

B C

N O

Z

D

P

A

9102

9103

9104

9101

9102

9103

9104

9105

9102

9103

9104

9105

9103

9104

9105

9104

9105

9105

Slide title 44 pt







footer area


› Source routing along an explicit labeled path

A path with Adjacency SID’s (strict encoding) MSD = 3

B C

N O

Z

D

P

A

9102

9103

9101

9102

9103

9102

9103 9103

9104

Slide title 44 pt







footer area


› #1 SID stack compression – Efficient path computation algorithms

›  Compressed SID stack that meets MSD limit

› #2 SID stack expansion – Instantiate a new SID stack at the anchor node, keeping initial stack within ingress’s MSD limits

›  Signaled thru Binding SID

Possible solutions: Control plane with exposure to SDNc is the right place to start!

Slide title 44 pt







footer area


› SR-LEA – SR paths Label Encoding Algorithm

› SR-LEA-A – SR paths Label Encoding Algorithm with global Adj-SID’s

Possible solutions - #1 SID stack compression

Label Encoding Algorithm for MPLS SegmentRouting

Rabah Guedrez⇤ Olivier Dugeon⇤ Samer Lahoud† Geraldine Texier‡⇤Orange Labs, Lannion, France,

[email protected], [email protected]†*IRISA/Universite de Rennes 1/Adopnet Team, Rennes, France, [email protected]

‡IRISA/Telecom Bretagne/Adopnet Team, Rennes, France, [email protected]

Abstract—Segment Routing is a new architecture that leveragesthe source routing mechanism to enhance packet forwarding innetworks. It is designed to operate over either an MPLS oran IPv6 control plane. SR-MPLS, its instantiation over MPLS,encodes a path as a stack of labels inserted in the packet headerby the ingress node. This overhead may violate the MaximumSID Depth (MSD), the equipment hardware limitation whichindicates the maximum number of labels an ingress node canpush onto the packet header. Currently, the MSD varies from3 to 5 depending on the equipment manufacturer. Therefore,the MSD value considerably limits the number of paths thatcan be implemented with SR-MPLS. The consequence may bean inefficient network resource utilization and may also lead tocongestion. We propose and analyze SR-LEA, an algorithm for anefficient path label encoding that takes advantage of the existingIGP shortest paths in the network. The output of SR-LEA is theminimum label stack to express SR-MPLS paths according tothe MSD constraint. Therefore, SR-LEA substantially slackensthe impact of MSD and restores the path diversity that MSDforbids in the network.

Index Terms—Segment Routing, MPLS, SR-MPLS, label stack,traffic engineering.

I. INTRODUCTION

Segment Routing (SR) is a new architecture standardized byIETF SPRING working group [1]. It can be instantiated overtwo existing data plane MPLS (SR-MPLS) [1] [2] and IPv6(SR-IPv6) [3]. In SR packet are forwarded using the sourcerouting mechanism: the path the packet has to go throughis encoded in its header. SR-MPLS is the central focus ofthe IETF working groups, mainly because of the importantimplications of service providers (SPs).

The major advantage of SR is that it eliminates the per-flowstates from the SP’s core routers. In fact, a path is directlyusable by any router; no prior setup/signalization is required,unlike MPLS-TE where a tunnel has to be signaled andmaintained using protocols such as the Resource ReservationProtocol Traffic Engineering (RSVP-TE). In SR, only theingress node has to maintain per-flow states. Also, SR archi-tecture adds extensions to already deployed IGP protocols:Open Shortest Path First (OSPF) [4], Intermediate System toIntermediate System (IS-IS) [5] and Border Gateway ProtocolLink State [6] to exchange SR information. Therefore, SR-MPLS revokes the need for a label distribution protocol suchas LDP or RSVP-TE.

A SR Path (SRP) is encoded as list of segments identifiers(SIDs), each SID associated with a data plane forwardinginstruction e.g., forward the packet down the IGP shortest pathor forward to a specific exit interface.

In the SR instantiation over the MPLS data plane (SR-MPLS), a SID is represented by a 20-bit label. The SID isprocessed using the three standard MPLS operations POP,PUSH, and SWAP. A SRP is encoded as a stack of labelsthat the ingress router pushes onto the packet header. In fact,pushing more than one was supported since the early versionof MPLS standards [7], the label stack has been used formultiple use cases: hierarchical tunnels, Layer 2 Virtual PrivateNetwork (L2VPN), and Layer 3 VPN. However, those usecases require a small number of labels, for example, a scenarioof L2VPN or L3VPN requires only simultaneously two labels:the tunnel’s label and VPN’s label. To take full advantageof SR’s potential, a router has to be able to push a largernumber of labels. Unfortunately, current hardware suffers fromphysical limitation of the number of labels that can be usedsimultaneously [8].

In fact, in order to achieve wire-speed packet processing,hardware vendors use Application-specific integrated circuit(ASIC)s. They are designed to perform specific tasks very effi-ciently compared to general purpose processors. Consequently,they are limited in the size and the type of the operations theycan perform. For example, the PUSH operation is implementedusing dedicated ASICs that limit the number of labels they canpush onto the packet header, this limitation in SR is knownas the Maximum SID Depth (MSD). Therefore, an efficientlabel encoding able to reduce the labels stack size is essentialto alleviate the MSD impact. In addition, reducing the labelstack saves space and enables to carry other types of labelssuch as the entropy labels [9].

In this paper, we propose two label encoding algorithmsfor SR-MPLS paths. Both algorithms compute the minimumnumber of labels to express a SRP. We evaluate theirperformances over several real-world network topologies.The results are presented in term of the average number oflabels to express a set network paths. In addition, we studytheir efficiency in alleviating the impact of the MSD limitation.

http://ieeexplore.ieee.org/document/7778603/ http://samer.lahoud.fr/pub-pdf/icin-17.pdf

Slide title 44 pt







footer area


› Analysis based on topologies available from Network Design Library – The result is the optimal set of paths to satisfy the demand matrix

›  V: the number of nodes ›  E: the number of links ›  D: number of demands in the demand matrix


as global segments it is the SR-LEA-A that computes theminimum label stack.

In SR-LEA-A, we suppose that the Adj-SIDs are advertisedas global segments, the resulting label stack is either smalleror equal to the SR-LEA’s one. Both algorithms share step 1detailed in Algorithm 1. In SR-LEA-A, as detailed by thepseudocode in Algorithm 2: a subpath of size > 3 followedby one of size = 2 are encoded using one label: the globalAdj-SID between the last node in the first path and the firstnode in the second one. Compared to SR-LEA, two labels areused to encode the two subpaths.

Algorithm 2 Efficient Label Encoding algorithm with globalAdj-SIDsSTEP 1 Same as for SR-LEASTEP 2

1: for i 1 To Size(A) do2: if length(A[i]) > 2 then3: if length(A[i+ 1]) == 2 then4: push(labelStack, GlobalAdjSID(

A[i][end], A[i+ 1][1]))5: p+ = 26: continue

7: end if8: push(labelStack,NodeSID(A[i][end]))9: else

10: push(labelStack,AdjSID(A[i]))11: end if12: end for

In the example described in Fig. 4, P3 advertisesits adjacency with P7 as the global SID 1037, thelist A contains the following subpaths: {(PE1, P2, P3),(P3, P7), (P7, P6, PE5)}. Accordingly, the two subpaths{(PE1, P2, P3), (P3, P7)} are encoded using the global Adj-SID P3 � P7 : 1037. Consequently, the label stack for thepath P is [1037, 1004]. At PE1 and P2, based on 1037 thepacket is forwarded down the shortest path to reach P3. AtP3, the top label 1037 is popped and the packet forwardedthrough the interface that connects P3 to P7. At P7, basedon the PE5’s Node-SID (i.e., 1005) the packet is forwardedthrough the shortest path to reach PE5.

V. SIMULATION RESULTS

In order to better evaluate the performance of the pro-posed algorithms, we experimented on several SNDlib networktopologies [17] [18]. To get a representative set of paths,for each topology, we consider a sample bandwidth demandmatrix D. As detailed in Table I, we solve the multicommodityflow problem [19]. The result is the optimal set of paths tosatisfy the demand matrix. The paths are then encoded usingthe strict Adj-SID, SR-LEA and SR-LEA-A.

The two proposed algorithms, compute the minimum labelstack to express a SRP. SR-LEA is used when the Adj-SIDs arelocal segments whilst SR-LEA-A is used when they are global.

Topology V E DGeant 22 36 431Albilene 12 18 131Brain 161 166 9045Germany50 50 80 1270Nobel-germany 17 26 248

TABLE I: V is the number of nodes. E the number of links.D: number of demands in the demand matrix.

0

1

2

3

4

5

6

7

Strict

SR-LEA

SR-LEA-A

Aver

age

labe

l sta

ck si

ze

Topologies

Fig. 6: Comparison of the average label stack size generatedusing a strict encoding, SR-LEA and SR-LEA-A algorithms.

The comparison is made between the strict encoding, the SR-LEA and the SR-LEA-A algorithms. For each topology, usingthe three encoding algorithms detailed previously, we computethe average label stack size and the percentage of networkpaths encoded with a label stack size MSD.

Fig. 6 illustrates the per-topology average label stack sizevariation depending on the topology and the encoding algo-rithm.

• We observe that the strict encoding always produces alarge label stack. This was expected because no optimiza-tion on the label stack size is performed, rather a one toone mapping of the physical links to the label stack. Wenote that for some paths the label stack noticeably reachesup to 14 labels.

• SR-LEA reduces the size of the label stack by 52% to65% compared to the strict encoding; the observed gainvaries depending on the network design and diameter.

• SR-LEA-A gives the best results. Notably, compared tothe strict encoding, the average label stack size is reducedby 57% to 67%.

The MSD corresponds to the maximum number of labels arouter can push onto packet header, it is a local characteristicof a router, it varies from one equipment vendor to another. Inan architecture where the path computation is delegated by theSR node to a centralized entity such as a SDN controller ora PCE. The node’s MSD is learned via the Path ComputationElement Protocol (PCEP) extensions for SR [8]. Hence, thislimitation is taken into consideration in the path computationprocess. This limitation makes long paths in the networkunusable. Consequently, it forces the network traffic to followonly short paths which cause inefficient traffic distribution orworse network congestion. For this study, we fixed the MSD

Slide title 44 pt







footer area


› Analysis based on topologies available from Network Design Library

› % of usable paths satisfying service requests with MSD = 5


Slide title 44 pt







footer area



› Average SID stack satisfying service requests


as global segments it is the SR-LEA-A that computes theminimum label stack.

In SR-LEA-A, we suppose that the Adj-SIDs are advertisedas global segments, the resulting label stack is either smalleror equal to the SR-LEA’s one. Both algorithms share step 1detailed in Algorithm 1. In SR-LEA-A, as detailed by thepseudocode in Algorithm 2: a subpath of size > 3 followedby one of size = 2 are encoded using one label: the globalAdj-SID between the last node in the first path and the firstnode in the second one. Compared to SR-LEA, two labels areused to encode the two subpaths.

Algorithm 2 Efficient Label Encoding algorithm with globalAdj-SIDsSTEP 1 Same as for SR-LEASTEP 2

1: for i 1 To Size(A) do2: if length(A[i]) > 2 then3: if length(A[i+ 1]) == 2 then4: push(labelStack, GlobalAdjSID(

A[i][end], A[i+ 1][1]))5: p+ = 26: continue

7: end if8: push(labelStack,NodeSID(A[i][end]))9: else

10: push(labelStack,AdjSID(A[i]))11: end if12: end for

In the example described in Fig. 4, P3 advertisesits adjacency with P7 as the global SID 1037, thelist A contains the following subpaths: {(PE1, P2, P3),(P3, P7), (P7, P6, PE5)}. Accordingly, the two subpaths{(PE1, P2, P3), (P3, P7)} are encoded using the global Adj-SID P3 � P7 : 1037. Consequently, the label stack for thepath P is [1037, 1004]. At PE1 and P2, based on 1037 thepacket is forwarded down the shortest path to reach P3. AtP3, the top label 1037 is popped and the packet forwardedthrough the interface that connects P3 to P7. At P7, basedon the PE5’s Node-SID (i.e., 1005) the packet is forwardedthrough the shortest path to reach PE5.

V. SIMULATION RESULTS

In order to better evaluate the performance of the pro-posed algorithms, we experimented on several SNDlib networktopologies [17] [18]. To get a representative set of paths,for each topology, we consider a sample bandwidth demandmatrix D. As detailed in Table I, we solve the multicommodityflow problem [19]. The result is the optimal set of paths tosatisfy the demand matrix. The paths are then encoded usingthe strict Adj-SID, SR-LEA and SR-LEA-A.

The two proposed algorithms, compute the minimum labelstack to express a SRP. SR-LEA is used when the Adj-SIDs arelocal segments whilst SR-LEA-A is used when they are global.

Topology V E DGeant 22 36 431Albilene 12 18 131Brain 161 166 9045Germany50 50 80 1270Nobel-germany 17 26 248

TABLE I: V is the number of nodes. E the number of links.D: number of demands in the demand matrix.

0

1

2

3

4

5

6

7

Strict

SR-LEA

SR-LEA-A

Aver

age

labe

l sta

ck si

ze

Topologies

Fig. 6: Comparison of the average label stack size generatedusing a strict encoding, SR-LEA and SR-LEA-A algorithms.

The comparison is made between the strict encoding, the SR-LEA and the SR-LEA-A algorithms. For each topology, usingthe three encoding algorithms detailed previously, we computethe average label stack size and the percentage of networkpaths encoded with a label stack size MSD.

Fig. 6 illustrates the per-topology average label stack sizevariation depending on the topology and the encoding algo-rithm.

• We observe that the strict encoding always produces alarge label stack. This was expected because no optimiza-tion on the label stack size is performed, rather a one toone mapping of the physical links to the label stack. Wenote that for some paths the label stack noticeably reachesup to 14 labels.

• SR-LEA reduces the size of the label stack by 52% to65% compared to the strict encoding; the observed gainvaries depending on the network design and diameter.

• SR-LEA-A gives the best results. Notably, compared tothe strict encoding, the average label stack size is reducedby 57% to 67%.

The MSD corresponds to the maximum number of labels arouter can push onto packet header, it is a local characteristicof a router, it varies from one equipment vendor to another. Inan architecture where the path computation is delegated by theSR node to a centralized entity such as a SDN controller ora PCE. The node’s MSD is learned via the Path ComputationElement Protocol (PCEP) extensions for SR [8]. Hence, thislimitation is taken into consideration in the path computationprocess. This limitation makes long paths in the networkunusable. Consequently, it forces the network traffic to followonly short paths which cause inefficient traffic distribution orworse network congestion. For this study, we fixed the MSD

Slide title 44 pt







footer area


Topologies

% P

aths

≤M

SD

0

20

40

60

80

100

120

Strict

SR-LEA

SR-LEA-A

Fig. 7: Paths expressed with a label stack size MSD

(MSD = 5).

to 5 labels, which is the value announced currently by themajor equipment vendors.

Fig. 7, illustrates the variation of the percentage of theuseable paths in each topology. With a strict encoding, thepercentage of useable paths can be very low e.g., 37% forGermany50 topology. Using SR-LEA, increases considerablythe amount of useable paths e.g., from 37% to 97% forGermany50 topology. However, encoding the label stackusing SR-LEA-A gives the best results, as it increases thenumber of usable paths from 37% to 99%, a gain of 2% to4% more than SR-LEA. We expect the difference to be moreconsiderable on topologies with bigger diameters.

We conclude that the proposed algorithms are very efficientin reducing the label stack size, also in minimize considerablythe impact of the MSD limitation. However, both algorithmsdo not completely eliminate the MSD problem, as we still havepaths that can not be expressed with a label stack smaller thanthe MSD.

VI. CONCLUSIONIn this work, we proposed two SR-MPLS paths label

encoding algorithms, namely SR-LEA and SR-LEA-A. Bothalgorithms compute the minimum label stack to express asegment routing path. Their performance has been evaluatedover real topologies. In addition, we prove that they areefficient in alleviating the impact of the MSD. For futurework, a PCE implementation of the proposed algorithms isunderdevelopment. We are considering the possibility to usethe two algorithms to encode Topology Independent Loop-FreeAlternate (TI-LFA) Fast Reroute post-convergence paths.

REFERENCES

[1] C. Filsfils, S. Previdi, B. Decraene, S. Litkowski, and R. Shakir,“Segment Routing Architecture,” Internet Engineering Task Force,Internet-Draft draft-ietf-spring-segment-routing-09, Jul. 2016, work inProgress. [Online]. Available: https://tools.ietf.org/html/draft-ietf-spring-segment-routing-09

[2] C. Filsfils, N. K. Nainar, C. Pignataro, J. C. Cardona, and P. Francois,“The segment routing architecture,” in 2015 IEEE Global Communica-tions Conference (GLOBECOM). IEEE, 2015, pp. 1–6.

[3] S. Previdi, C. Filsfils, B. Field, I. Leung, J. Linkova, E. Aries,T. Kosugi, E. Vyncke, and D. Lebrun, “IPv6 Segment RoutingHeader (SRH),” Internet Engineering Task Force, Internet-Draft draft-ietf-6man-segment-routing-header-01, Mar. 2016, work in Progress.[Online]. Available: https://tools.ietf.org/html/draft-ietf-6man-segment-routing-header-01

[4] P. Psenak, S. Previdi, C. Filsfils, W. Henderickx, J. Tantsura, H. Gredler,and R. Shakir, “OSPF Extensions for Segment Routing,” InternetEngineering Task Force, Internet-Draft draft-ietf-ospf-segment-routing-extensions-08, Apr. 2016, work in Progress. [Online]. Available:https://tools.ietf.org/html/draft-ietf-ospf-segment-routing-extensions-08

[5] S. Previdi, C. Filsfils, A. Bashandy, S. Litkowski, J. Tantsura,B. Decraene, and H. Gredler, “IS-IS Extensions for Segment Routing,”Internet Engineering Task Force, Internet-Draft draft-ietf-isis-segment-routing-extensions-06, Mar. 2016, work in Progress. [Online]. Available:https://tools.ietf.org/html/draft-ietf-isis-segment-routing-extensions-06

[6] S. Previdi, P. Psenak, C. Filsfils, H. Gredler, M. Chen, andJ. Tantsura, “BGP Link-State extensions for Segment Routing,”Internet Engineering Task Force, Internet-Draft draft-gredler-idr-bgp-ls-segment-routing-ext-01, Dec. 2015, work in Progress.[Online]. Available: https://tools.ietf.org/html/draft-gredler-idr-bgp-ls-segment-routing-ext-01

[7] Y. Rekhter, A. Conta, G. Fedorkow, E. Rosen, D. Farinacci, andT. Li, “MPLS Label Stack Encoding,” RFC 3032, Mar. 2013. [Online].Available: https://rfc-editor.org/rfc/rfc3032.txt

[8] S. Sivabalan, J. Medved, C. Filsfils, V. Lopez, J. Tantsura,W. Henderickx, E. Crabbe, and J. Hardwick, “PCEP Extensions forSegment Routing,” Internet Engineering Task Force, Internet-Draft draft-ietf-pce-segment-routing-07, Mar. 2016, work in Progress. [Online].Available: https://tools.ietf.org/html/draft-ietf-pce-segment-routing-07

[9] K. Kompella, S. Sivabalan, S. Litkowski, R. Shakir, S. Kini,and [email protected], “Entropy labels for source routed tunnelswith label stacks,” Internet Engineering Task Force, Internet-Draft draft-ietf-mpls-spring-entropy-label-03, Apr. 2016, work inProgress. [Online]. Available: https://tools.ietf.org/html/draft-ietf-mpls-spring-entropy-label-03

[10] A. Giorgetti, P. Castoldi, F. Cugini, J. Nijhof, F. Lazzeri, and G. Bruno,“Path encoding in segment routing,” in 2015 IEEE Global Communica-tions Conference (GLOBECOM). IEEE, 2015, pp. 1–6.

[11] F. Lazzeri, G. Bruno, J. Nijhof, A. Giorgetti, and P. Castoldi, “Efficientlabel encoding in segment-routing enabled optical networks,” in OpticalNetwork Design and Modeling (ONDM), 2015 International Conferenceon. IEEE, 2015, pp. 34–38.

[12] R. Geib, C. Filsfils, C. Pignataro, and N. Kumar, “AScalable and Topology-Aware MPLS Dataplane Monitoring System,”Internet Engineering Task Force, Internet-Draft draft-ietf-spring-oam-usecase-03, Apr. 2016, work in Progress. [Online]. Available:https://tools.ietf.org/html/draft-ietf-spring-oam-usecase-03

[13] A. Sgambelluri, A. Giorgetti, F. Cugini, G. Bruno, F. Lazzeri, andP. Castoldi, “First demonstration of sdn-based segment routing in multi-layer networks,” in Optical Fiber Communications Conference andExhibition (OFC), 2015. IEEE, 2015, pp. 1–3.

[14] L. Davoli, L. Veltri, P. L. Ventre, G. Siracusano, and S. Salsano, “Trafficengineering with segment routing: Sdn-based architectural design andopen source implementation,” in 2015 Fourth European Workshop onSoftware Defined Networks. IEEE, 2015, pp. 111–112.

[15] J. Medved, I. Minei, E. Crabbe, and R. Varga, “PCEP Extensionsfor Stateful PCE,” Internet Engineering Task Force, Internet-Draftdraft-ietf-pce-stateful-pce-14, Mar. 2016, work in Progress. [Online].Available: https://tools.ietf.org/html/draft-ietf-pce-stateful-pce-14

[16] A. Sgambelluri, F. Paolucci, A. Giorgetti, F. Cugini, and P. Castoldi,“Sdn and pce implementations for segment routing,” in Networks andOptical Communications-(NOC), 2015 20th European Conference on.IEEE, 2015, pp. 1–4.

[17] S. Orlowski, M. Pioro, A. Tomaszewski, and R. Wessaly,“SNDlib 1.0–Survivable Network Design Library,” in Proceedingsof the 3rd International Network Optimization Conference(INOC 2007), Spa, Belgium, April 2007, http://sndlib.zib.de,extended version accepted in Networks, 2009. [Online]. Available:http://www.zib.de/orlowski/Paper/OrlowskiPioroTomaszewskiWessaely2007-SNDlib-INOC.pdf.gz

[18] ——, “SNDlib 1.0–Survivable Network Design Library,” Networks,vol. 55, no. 3, pp. 276–286, 2010. [Online]. Available:http://www3.interscience.wiley.com/journal/122653325/abstract

[19] M. Pioro and D. Medhi, Routing, flow, and capacity design in commu-nication and computer networks. Elsevier, 2004.


› % of usable paths satisfying service requests with and without compression


Slide title 44 pt







footer area


› Node O (anchor for Binding SID 25001) expands 25001 into new SID stack {9104,9105}

Possible solutions - #2 SID stack expansion, MSD = 4

B C

N O

Z

D

P

A

9102

9103

9104

9101

9102

9103

25001

9102

9103

25001

9103

25001

9104

9105

9105

Binding SID

Slide title 44 pt







footer area


›  IETF MSD drafts (ready for the Working Group Last Call) – OSPF

›  draft-ietf-ospf-segment-routing-msd – ISIS

›  draft-ietf-isis-segment-routing-msd – BGP-LS

›  draft-ietf-idr-bgp-ls-segment-routing-msd

› PCEP – Binding SID setup

›  draft-sivabalan-pce-binding-label-sid

›  Common IANA registry for MSD types for OSPF/ISIS/BGP allows simplified and quick addition/adoption of new types (IPv6, recirculation, entropy, etc)

Signaling (standardization)

Slide title 44 pt







footer area


› HW limitations are impairing service agility – TTM for a new ASIC is around 2 years

›  Innovation in SW provides tangible results – Work in IETF ensures - the solution is technically sound and can interoperate

› Get your vendors to implement it J – First implementations are there (FRR, Cisco (planned))

Conclusions

Questions

Slide title 44 pt







footer area


Thank you!

Date post:	09-Mar-2018
Category:	Documents
Upload:	tranlien
View:	216 times
Download:	3 times

1440-1500 20171002 Tantsura The Critical Role v1 · PDF...

Documents